Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
FreeBSD/Linux Kernel Cross Reference
sys/netinet/tcp_syncache.c
Version:
- FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
SearchContext: - none - 3 - 10
1 /*- 2 * Copyright (c) 2001 Networks Associates Technologies, Inc. 3 * All rights reserved. 4 * 5 * This software was developed for the FreeBSD Project by Jonathan Lemon 6 * and NAI Labs, the Security Research Division of Network Associates, Inc. 7 * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the 8 * DARPA CHATS research program. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. The name of the author may not be used to endorse or promote 19 * products derived from this software without specific prior written 20 * permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * $FreeBSD$ 35 */ 36 37 #include "opt_inet.h" 38 #include "opt_inet6.h" 39 #include "opt_ipsec.h" 40 #include "opt_random_ip_id.h" 41 42 #include <sys/param.h> 43 #include <sys/systm.h> 44 #include <sys/kernel.h> 45 #include <sys/sysctl.h> 46 #include <sys/malloc.h> 47 #include <sys/mbuf.h> 48 #include <sys/md5.h> 49 #include <sys/proc.h> /* for proc0 declaration */ 50 #include <sys/random.h> 51 #include <sys/socket.h> 52 #include <sys/socketvar.h> 53 54 #include <net/if.h> 55 #include <net/route.h> 56 57 #include <netinet/in.h> 58 #include <netinet/in_systm.h> 59 #include <netinet/ip.h> 60 #include <netinet/in_var.h> 61 #include <netinet/in_pcb.h> 62 #include <netinet/ip_var.h> 63 #ifdef INET6 64 #include <netinet/ip6.h> 65 #include <netinet/icmp6.h> 66 #include <netinet6/nd6.h> 67 #include <netinet6/ip6_var.h> 68 #include <netinet6/in6_pcb.h> 69 #endif 70 #include <netinet/tcp.h> 71 #include <netinet/tcp_fsm.h> 72 #include <netinet/tcp_seq.h> 73 #include <netinet/tcp_timer.h> 74 #include <netinet/tcp_var.h> 75 #ifdef INET6 76 #include <netinet6/tcp6_var.h> 77 #endif 78 79 #ifdef IPSEC 80 #include <netinet6/ipsec.h> 81 #ifdef INET6 82 #include <netinet6/ipsec6.h> 83 #endif 84 #include <netkey/key.h> 85 #endif /*IPSEC*/ 86 87 #ifdef FAST_IPSEC 88 #include <netipsec/ipsec.h> 89 #ifdef INET6 90 #include <netipsec/ipsec6.h> 91 #endif 92 #include <netipsec/key.h> 93 #define IPSEC 94 #endif /*FAST_IPSEC*/ 95 96 #include <machine/in_cksum.h> 97 #include <vm/vm_zone.h> 98 99 static int tcp_syncookies = 1; 100 SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_RW, 101 &tcp_syncookies, 0, 102 "Use TCP SYN cookies if the syncache overflows"); 103 104 static void syncache_drop(struct syncache *, struct syncache_head *); 105 static void syncache_free(struct syncache *); 106 static void syncache_insert(struct syncache *, struct syncache_head *); 107 struct syncache *syncache_lookup(struct in_conninfo *, struct syncache_head **); 108 static int syncache_respond(struct syncache *, struct mbuf *); 109 static struct socket *syncache_socket(struct syncache *, struct socket *); 110 static void syncache_timer(void *); 111 static u_int32_t syncookie_generate(struct syncache *, u_int32_t *); 112 static struct syncache *syncookie_lookup(struct in_conninfo *, 113 struct tcphdr *, struct socket *); 114 115 /* 116 * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies. 117 * 3 retransmits corresponds to a timeout of (1 + 2 + 4 + 8 == 15) seconds, 118 * the odds are that the user has given up attempting to connect by then. 119 */ 120 #define SYNCACHE_MAXREXMTS 3 121 122 /* Arbitrary values */ 123 #define TCP_SYNCACHE_HASHSIZE 512 124 #define TCP_SYNCACHE_BUCKETLIMIT 30 125 126 struct tcp_syncache { 127 struct syncache_head *hashbase; 128 struct vm_zone *zone; 129 u_int hashsize; 130 u_int hashmask; 131 u_int bucket_limit; 132 u_int cache_count; 133 u_int cache_limit; 134 u_int rexmt_limit; 135 u_int hash_secret; 136 u_int next_reseed; 137 TAILQ_HEAD(, syncache) timerq[SYNCACHE_MAXREXMTS + 1]; 138 struct callout tt_timerq[SYNCACHE_MAXREXMTS + 1]; 139 }; 140 static struct tcp_syncache tcp_syncache; 141 142 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache"); 143 144 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_RD, 145 &tcp_syncache.bucket_limit, 0, "Per-bucket hash limit for syncache"); 146 147 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_RD, 148 &tcp_syncache.cache_limit, 0, "Overall entry limit for syncache"); 149 150 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_RD, 151 &tcp_syncache.cache_count, 0, "Current number of entries in syncache"); 152 153 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_RD, 154 &tcp_syncache.hashsize, 0, "Size of TCP syncache hashtable"); 155 156 SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_RW, 157 &tcp_syncache.rexmt_limit, 0, "Limit on SYN/ACK retransmissions"); 158 159 static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache"); 160 161 #define SYNCACHE_HASH(inc, mask) \ 162 ((tcp_syncache.hash_secret ^ \ 163 (inc)->inc_faddr.s_addr ^ \ 164 ((inc)->inc_faddr.s_addr >> 16) ^ \ 165 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 166 167 #define SYNCACHE_HASH6(inc, mask) \ 168 ((tcp_syncache.hash_secret ^ \ 169 (inc)->inc6_faddr.s6_addr32[0] ^ \ 170 (inc)->inc6_faddr.s6_addr32[3] ^ \ 171 (inc)->inc_fport ^ (inc)->inc_lport) & mask) 172 173 #define ENDPTS_EQ(a, b) ( \ 174 (a)->ie_fport == (b)->ie_fport && \ 175 (a)->ie_lport == (b)->ie_lport && \ 176 (a)->ie_faddr.s_addr == (b)->ie_faddr.s_addr && \ 177 (a)->ie_laddr.s_addr == (b)->ie_laddr.s_addr \ 178 ) 179 180 #define ENDPTS6_EQ(a, b) (memcmp(a, b, sizeof(*a)) == 0) 181 182 #define SYNCACHE_TIMEOUT(sc, slot) do { \ 183 sc->sc_rxtslot = slot; \ 184 sc->sc_rxttime = ticks + TCPTV_RTOBASE * tcp_backoff[slot]; \ 185 TAILQ_INSERT_TAIL(&tcp_syncache.timerq[slot], sc, sc_timerq); \ 186 if (!callout_active(&tcp_syncache.tt_timerq[slot])) \ 187 callout_reset(&tcp_syncache.tt_timerq[slot], \ 188 TCPTV_RTOBASE * tcp_backoff[slot], \ 189 syncache_timer, (void *)((intptr_t)slot)); \ 190 } while (0) 191 192 static void 193 syncache_free(struct syncache *sc) 194 { 195 struct rtentry *rt; 196 197 if (sc->sc_ipopts) 198 (void) m_free(sc->sc_ipopts); 199 #ifdef INET6 200 if (sc->sc_inc.inc_isipv6) 201 rt = sc->sc_route6.ro_rt; 202 else 203 #endif 204 rt = sc->sc_route.ro_rt; 205 if (rt != NULL) { 206 /* 207 * If this is the only reference to a protocol cloned 208 * route, remove it immediately. 209 */ 210 if (rt->rt_flags & RTF_WASCLONED && 211 (sc->sc_flags & SCF_KEEPROUTE) == 0 && 212 rt->rt_refcnt == 1) 213 rtrequest(RTM_DELETE, rt_key(rt), 214 rt->rt_gateway, rt_mask(rt), 215 rt->rt_flags, NULL); 216 RTFREE(rt); 217 } 218 zfree(tcp_syncache.zone, sc); 219 } 220 221 void 222 syncache_init(void) 223 { 224 int i; 225 226 tcp_syncache.cache_count = 0; 227 tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; 228 tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT; 229 tcp_syncache.cache_limit = 230 tcp_syncache.hashsize * tcp_syncache.bucket_limit; 231 tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS; 232 tcp_syncache.next_reseed = 0; 233 tcp_syncache.hash_secret = arc4random(); 234 235 TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize", 236 &tcp_syncache.hashsize); 237 TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit", 238 &tcp_syncache.cache_limit); 239 TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit", 240 &tcp_syncache.bucket_limit); 241 if (!powerof2(tcp_syncache.hashsize)) { 242 printf("WARNING: syncache hash size is not a power of 2.\n"); 243 tcp_syncache.hashsize = 512; /* safe default */ 244 } 245 tcp_syncache.hashmask = tcp_syncache.hashsize - 1; 246 247 /* Allocate the hash table. */ 248 MALLOC(tcp_syncache.hashbase, struct syncache_head *, 249 tcp_syncache.hashsize * sizeof(struct syncache_head), 250 M_SYNCACHE, M_WAITOK); 251 252 /* Initialize the hash buckets. */ 253 for (i = 0; i < tcp_syncache.hashsize; i++) { 254 TAILQ_INIT(&tcp_syncache.hashbase[i].sch_bucket); 255 tcp_syncache.hashbase[i].sch_length = 0; 256 } 257 258 /* Initialize the timer queues. */ 259 for (i = 0; i <= SYNCACHE_MAXREXMTS; i++) { 260 TAILQ_INIT(&tcp_syncache.timerq[i]); 261 callout_init(&tcp_syncache.tt_timerq[i]); 262 } 263 264 /* 265 * Allocate the syncache entries. Allow the zone to allocate one 266 * more entry than cache limit, so a new entry can bump out an 267 * older one. 268 */ 269 tcp_syncache.cache_limit -= 1; 270 tcp_syncache.zone = zinit("syncache", sizeof(struct syncache), 271 tcp_syncache.cache_limit, ZONE_INTERRUPT, 0); 272 } 273 274 static void 275 syncache_insert(sc, sch) 276 struct syncache *sc; 277 struct syncache_head *sch; 278 { 279 struct syncache *sc2; 280 int s, i; 281 282 /* 283 * Make sure that we don't overflow the per-bucket 284 * limit or the total cache size limit. 285 */ 286 s = splnet(); 287 if (sch->sch_length >= tcp_syncache.bucket_limit) { 288 /* 289 * The bucket is full, toss the oldest element. 290 */ 291 sc2 = TAILQ_FIRST(&sch->sch_bucket); 292 sc2->sc_tp->ts_recent = ticks; 293 syncache_drop(sc2, sch); 294 tcpstat.tcps_sc_bucketoverflow++; 295 } else if (tcp_syncache.cache_count >= tcp_syncache.cache_limit) { 296 /* 297 * The cache is full. Toss the oldest entry in the 298 * entire cache. This is the front entry in the 299 * first non-empty timer queue with the largest 300 * timeout value. 301 */ 302 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) { 303 sc2 = TAILQ_FIRST(&tcp_syncache.timerq[i]); 304 if (sc2 != NULL) 305 break; 306 } 307 sc2->sc_tp->ts_recent = ticks; 308 syncache_drop(sc2, NULL); 309 tcpstat.tcps_sc_cacheoverflow++; 310 } 311 312 /* Initialize the entry's timer. */ 313 SYNCACHE_TIMEOUT(sc, 0); 314 315 /* Put it into the bucket. */ 316 TAILQ_INSERT_TAIL(&sch->sch_bucket, sc, sc_hash); 317 sch->sch_length++; 318 tcp_syncache.cache_count++; 319 tcpstat.tcps_sc_added++; 320 splx(s); 321 } 322 323 static void 324 syncache_drop(sc, sch) 325 struct syncache *sc; 326 struct syncache_head *sch; 327 { 328 int s; 329 330 if (sch == NULL) { 331 #ifdef INET6 332 if (sc->sc_inc.inc_isipv6) { 333 sch = &tcp_syncache.hashbase[ 334 SYNCACHE_HASH6(&sc->sc_inc, tcp_syncache.hashmask)]; 335 } else 336 #endif 337 { 338 sch = &tcp_syncache.hashbase[ 339 SYNCACHE_HASH(&sc->sc_inc, tcp_syncache.hashmask)]; 340 } 341 } 342 343 s = splnet(); 344 345 TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); 346 sch->sch_length--; 347 tcp_syncache.cache_count--; 348 349 TAILQ_REMOVE(&tcp_syncache.timerq[sc->sc_rxtslot], sc, sc_timerq); 350 if (TAILQ_EMPTY(&tcp_syncache.timerq[sc->sc_rxtslot])) 351 callout_stop(&tcp_syncache.tt_timerq[sc->sc_rxtslot]); 352 splx(s); 353 354 syncache_free(sc); 355 } 356 357 /* 358 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 359 * If we have retransmitted an entry the maximum number of times, expire it. 360 */ 361 static void 362 syncache_timer(xslot) 363 void *xslot; 364 { 365 intptr_t slot = (intptr_t)xslot; 366 struct syncache *sc, *nsc; 367 struct inpcb *inp; 368 int s; 369 370 s = splnet(); 371 if (callout_pending(&tcp_syncache.tt_timerq[slot]) || 372 !callout_active(&tcp_syncache.tt_timerq[slot])) { 373 splx(s); 374 return; 375 } 376 callout_deactivate(&tcp_syncache.tt_timerq[slot]); 377 378 nsc = TAILQ_FIRST(&tcp_syncache.timerq[slot]); 379 while (nsc != NULL) { 380 if (ticks < nsc->sc_rxttime) 381 break; 382 sc = nsc; 383 inp = sc->sc_tp->t_inpcb; 384 if (slot == SYNCACHE_MAXREXMTS || 385 slot >= tcp_syncache.rexmt_limit || 386 inp->inp_gencnt != sc->sc_inp_gencnt) { 387 nsc = TAILQ_NEXT(sc, sc_timerq); 388 syncache_drop(sc, NULL); 389 tcpstat.tcps_sc_stale++; 390 continue; 391 } 392 /* 393 * syncache_respond() may call back into the syncache to 394 * to modify another entry, so do not obtain the next 395 * entry on the timer chain until it has completed. 396 */ 397 (void) syncache_respond(sc, NULL); 398 nsc = TAILQ_NEXT(sc, sc_timerq); 399 tcpstat.tcps_sc_retransmitted++; 400 TAILQ_REMOVE(&tcp_syncache.timerq[slot], sc, sc_timerq); 401 SYNCACHE_TIMEOUT(sc, slot + 1); 402 } 403 if (nsc != NULL) 404 callout_reset(&tcp_syncache.tt_timerq[slot], 405 nsc->sc_rxttime - ticks, syncache_timer, (void *)(slot)); 406 splx(s); 407 } 408 409 /* 410 * Find an entry in the syncache. 411 */ 412 struct syncache * 413 syncache_lookup(inc, schp) 414 struct in_conninfo *inc; 415 struct syncache_head **schp; 416 { 417 struct syncache *sc; 418 struct syncache_head *sch; 419 int s; 420 421 #ifdef INET6 422 if (inc->inc_isipv6) { 423 sch = &tcp_syncache.hashbase[ 424 SYNCACHE_HASH6(inc, tcp_syncache.hashmask)]; 425 *schp = sch; 426 s = splnet(); 427 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 428 if (ENDPTS6_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) { 429 splx(s); 430 return (sc); 431 } 432 } 433 splx(s); 434 } else 435 #endif 436 { 437 sch = &tcp_syncache.hashbase[ 438 SYNCACHE_HASH(inc, tcp_syncache.hashmask)]; 439 *schp = sch; 440 s = splnet(); 441 TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { 442 #ifdef INET6 443 if (sc->sc_inc.inc_isipv6) 444 continue; 445 #endif 446 if (ENDPTS_EQ(&inc->inc_ie, &sc->sc_inc.inc_ie)) { 447 splx(s); 448 return (sc); 449 } 450 } 451 splx(s); 452 } 453 return (NULL); 454 } 455 456 /* 457 * This function is called when we get a RST for a 458 * non-existent connection, so that we can see if the 459 * connection is in the syn cache. If it is, zap it. 460 */ 461 void 462 syncache_chkrst(inc, th) 463 struct in_conninfo *inc; 464 struct tcphdr *th; 465 { 466 struct syncache *sc; 467 struct syncache_head *sch; 468 469 sc = syncache_lookup(inc, &sch); 470 if (sc == NULL) 471 return; 472 /* 473 * If the RST bit is set, check the sequence number to see 474 * if this is a valid reset segment. 475 * RFC 793 page 37: 476 * In all states except SYN-SENT, all reset (RST) segments 477 * are validated by checking their SEQ-fields. A reset is 478 * valid if its sequence number is in the window. 479 * 480 * The sequence number in the reset segment is normally an 481 * echo of our outgoing acknowlegement numbers, but some hosts 482 * send a reset with the sequence number at the rightmost edge 483 * of our receive window, and we have to handle this case. 484 */ 485 if (SEQ_GEQ(th->th_seq, sc->sc_irs) && 486 SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { 487 syncache_drop(sc, sch); 488 tcpstat.tcps_sc_reset++; 489 } 490 } 491 492 void 493 syncache_badack(inc) 494 struct in_conninfo *inc; 495 { 496 struct syncache *sc; 497 struct syncache_head *sch; 498 499 sc = syncache_lookup(inc, &sch); 500 if (sc != NULL) { 501 syncache_drop(sc, sch); 502 tcpstat.tcps_sc_badack++; 503 } 504 } 505 506 void 507 syncache_unreach(inc, th) 508 struct in_conninfo *inc; 509 struct tcphdr *th; 510 { 511 struct syncache *sc; 512 struct syncache_head *sch; 513 514 /* we are called at splnet() here */ 515 sc = syncache_lookup(inc, &sch); 516 if (sc == NULL) 517 return; 518 519 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 520 if (ntohl(th->th_seq) != sc->sc_iss) 521 return; 522 523 /* 524 * If we've rertransmitted 3 times and this is our second error, 525 * we remove the entry. Otherwise, we allow it to continue on. 526 * This prevents us from incorrectly nuking an entry during a 527 * spurious network outage. 528 * 529 * See tcp_notify(). 530 */ 531 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtslot < 3) { 532 sc->sc_flags |= SCF_UNREACH; 533 return; 534 } 535 syncache_drop(sc, sch); 536 tcpstat.tcps_sc_unreach++; 537 } 538 539 /* 540 * Build a new TCP socket structure from a syncache entry. 541 */ 542 static struct socket * 543 syncache_socket(sc, lso) 544 struct syncache *sc; 545 struct socket *lso; 546 { 547 struct inpcb *inp = NULL; 548 struct socket *so; 549 struct tcpcb *tp; 550 551 /* 552 * Ok, create the full blown connection, and set things up 553 * as they would have been set up if we had created the 554 * connection when the SYN arrived. If we can't create 555 * the connection, abort it. 556 */ 557 so = sonewconn(lso, SS_ISCONNECTED); 558 if (so == NULL) { 559 /* 560 * Drop the connection; we will send a RST if the peer 561 * retransmits the ACK, 562 */ 563 tcpstat.tcps_listendrop++; 564 goto abort; 565 } 566 567 inp = sotoinpcb(so); 568 569 /* 570 * Insert new socket into hash list. 571 */ 572 inp->inp_inc.inc_isipv6 = sc->sc_inc.inc_isipv6; 573 #ifdef INET6 574 if (sc->sc_inc.inc_isipv6) { 575 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 576 } else { 577 inp->inp_vflag &= ~INP_IPV6; 578 inp->inp_vflag |= INP_IPV4; 579 #endif 580 inp->inp_laddr = sc->sc_inc.inc_laddr; 581 #ifdef INET6 582 } 583 #endif 584 inp->inp_lport = sc->sc_inc.inc_lport; 585 if (in_pcbinshash(inp) != 0) { 586 /* 587 * Undo the assignments above if we failed to 588 * put the PCB on the hash lists. 589 */ 590 #ifdef INET6 591 if (sc->sc_inc.inc_isipv6) 592 inp->in6p_laddr = in6addr_any; 593 else 594 #endif 595 inp->inp_laddr.s_addr = INADDR_ANY; 596 inp->inp_lport = 0; 597 goto abort; 598 } 599 #ifdef IPSEC 600 /* copy old policy into new socket's */ 601 if (ipsec_copy_policy(sotoinpcb(lso)->inp_sp, inp->inp_sp)) 602 printf("syncache_expand: could not copy policy\n"); 603 #endif 604 #ifdef INET6 605 if (sc->sc_inc.inc_isipv6) { 606 struct inpcb *oinp = sotoinpcb(lso); 607 struct in6_addr laddr6; 608 struct sockaddr_in6 *sin6; 609 /* 610 * Inherit socket options from the listening socket. 611 * Note that in6p_inputopts are not (and should not be) 612 * copied, since it stores previously received options and is 613 * used to detect if each new option is different than the 614 * previous one and hence should be passed to a user. 615 * If we copied in6p_inputopts, a user would not be able to 616 * receive options just after calling the accept system call. 617 */ 618 inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS; 619 if (oinp->in6p_outputopts) 620 inp->in6p_outputopts = 621 ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT); 622 inp->in6p_route = sc->sc_route6; 623 sc->sc_route6.ro_rt = NULL; 624 625 MALLOC(sin6, struct sockaddr_in6 *, sizeof *sin6, 626 M_SONAME, M_NOWAIT | M_ZERO); 627 if (sin6 == NULL) 628 goto abort; 629 sin6->sin6_family = AF_INET6; 630 sin6->sin6_len = sizeof(*sin6); 631 sin6->sin6_addr = sc->sc_inc.inc6_faddr; 632 sin6->sin6_port = sc->sc_inc.inc_fport; 633 laddr6 = inp->in6p_laddr; 634 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 635 inp->in6p_laddr = sc->sc_inc.inc6_laddr; 636 if (in6_pcbconnect(inp, (struct sockaddr *)sin6, &proc0)) { 637 inp->in6p_laddr = laddr6; 638 FREE(sin6, M_SONAME); 639 goto abort; 640 } 641 FREE(sin6, M_SONAME); 642 /* Override flowlabel from in6_pcbconnect. */ 643 inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK; 644 inp->in6p_flowinfo |= sc->sc_flowlabel; 645 } else 646 #endif 647 { 648 struct in_addr laddr; 649 struct sockaddr_in *sin; 650 651 inp->inp_options = ip_srcroute(); 652 if (inp->inp_options == NULL) { 653 inp->inp_options = sc->sc_ipopts; 654 sc->sc_ipopts = NULL; 655 } 656 inp->inp_route = sc->sc_route; 657 sc->sc_route.ro_rt = NULL; 658 659 MALLOC(sin, struct sockaddr_in *, sizeof *sin, 660 M_SONAME, M_NOWAIT | M_ZERO); 661 if (sin == NULL) 662 goto abort; 663 sin->sin_family = AF_INET; 664 sin->sin_len = sizeof(*sin); 665 sin->sin_addr = sc->sc_inc.inc_faddr; 666 sin->sin_port = sc->sc_inc.inc_fport; 667 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 668 laddr = inp->inp_laddr; 669 if (inp->inp_laddr.s_addr == INADDR_ANY) 670 inp->inp_laddr = sc->sc_inc.inc_laddr; 671 if (in_pcbconnect(inp, (struct sockaddr *)sin, &proc0)) { 672 inp->inp_laddr = laddr; 673 FREE(sin, M_SONAME); 674 goto abort; 675 } 676 FREE(sin, M_SONAME); 677 } 678 679 tp = intotcpcb(inp); 680 tp->t_state = TCPS_SYN_RECEIVED; 681 tp->iss = sc->sc_iss; 682 tp->irs = sc->sc_irs; 683 tcp_rcvseqinit(tp); 684 tcp_sendseqinit(tp); 685 tp->snd_wl1 = sc->sc_irs; 686 tp->rcv_up = sc->sc_irs + 1; 687 tp->rcv_wnd = sc->sc_wnd; 688 tp->rcv_adv += tp->rcv_wnd; 689 690 tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY); 691 if (sc->sc_flags & SCF_NOOPT) 692 tp->t_flags |= TF_NOOPT; 693 if (sc->sc_flags & SCF_WINSCALE) { 694 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 695 tp->requested_s_scale = sc->sc_requested_s_scale; 696 tp->request_r_scale = sc->sc_request_r_scale; 697 } 698 if (sc->sc_flags & SCF_TIMESTAMP) { 699 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 700 tp->ts_recent = sc->sc_tsrecent; 701 tp->ts_recent_age = ticks; 702 } 703 if (sc->sc_flags & SCF_CC) { 704 /* 705 * Initialization of the tcpcb for transaction; 706 * set SND.WND = SEG.WND, 707 * initialize CCsend and CCrecv. 708 */ 709 tp->t_flags |= TF_REQ_CC|TF_RCVD_CC; 710 tp->cc_send = sc->sc_cc_send; 711 tp->cc_recv = sc->sc_cc_recv; 712 } 713 #ifdef TCP_SIGNATURE 714 if (sc->sc_flags & SCF_SIGNATURE) 715 tp->t_flags |= TF_SIGNATURE; 716 #endif 717 718 tcp_mss(tp, sc->sc_peer_mss); 719 720 /* 721 * If the SYN,ACK was retransmitted, reset cwnd to 1 segment. 722 */ 723 if (sc->sc_rxtslot != 0) 724 tp->snd_cwnd = tp->t_maxseg; 725 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp); 726 727 tcpstat.tcps_accepts++; 728 return (so); 729 730 abort: 731 if (so != NULL) 732 (void) soabort(so); 733 return (NULL); 734 } 735 736 /* 737 * This function gets called when we receive an ACK for a 738 * socket in the LISTEN state. We look up the connection 739 * in the syncache, and if its there, we pull it out of 740 * the cache and turn it into a full-blown connection in 741 * the SYN-RECEIVED state. 742 */ 743 int 744 syncache_expand(inc, th, sop, m) 745 struct in_conninfo *inc; 746 struct tcphdr *th; 747 struct socket **sop; 748 struct mbuf *m; 749 { 750 struct syncache *sc; 751 struct syncache_head *sch; 752 struct socket *so; 753 754 sc = syncache_lookup(inc, &sch); 755 if (sc == NULL) { 756 /* 757 * There is no syncache entry, so see if this ACK is 758 * a returning syncookie. To do this, first: 759 * A. See if this socket has had a syncache entry dropped in 760 * the past. We don't want to accept a bogus syncookie 761 * if we've never received a SYN. 762 * B. check that the syncookie is valid. If it is, then 763 * cobble up a fake syncache entry, and return. 764 */ 765 if (!tcp_syncookies) 766 return (0); 767 sc = syncookie_lookup(inc, th, *sop); 768 if (sc == NULL) 769 return (0); 770 sch = NULL; 771 tcpstat.tcps_sc_recvcookie++; 772 } 773 774 /* 775 * If seg contains an ACK, but not for our SYN/ACK, send a RST. 776 */ 777 if (th->th_ack != sc->sc_iss + 1) 778 return (0); 779 780 so = syncache_socket(sc, *sop); 781 if (so == NULL) { 782 #if 0 783 resetandabort: 784 /* XXXjlemon check this - is this correct? */ 785 (void) tcp_respond(NULL, m, m, th, 786 th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK); 787 #endif 788 m_freem(m); /* XXX only needed for above */ 789 tcpstat.tcps_sc_aborted++; 790 } else { 791 sc->sc_flags |= SCF_KEEPROUTE; 792 tcpstat.tcps_sc_completed++; 793 } 794 if (sch == NULL) 795 syncache_free(sc); 796 else 797 syncache_drop(sc, sch); 798 *sop = so; 799 return (1); 800 } 801 802 /* 803 * Given a LISTEN socket and an inbound SYN request, add 804 * this to the syn cache, and send back a segment: 805 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 806 * to the source. 807 * 808 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 809 * Doing so would require that we hold onto the data and deliver it 810 * to the application. However, if we are the target of a SYN-flood 811 * DoS attack, an attacker could send data which would eventually 812 * consume all available buffer space if it were ACKed. By not ACKing 813 * the data, we avoid this DoS scenario. 814 */ 815 int 816 syncache_add(inc, to, th, sop, m) 817 struct in_conninfo *inc; 818 struct tcpopt *to; 819 struct tcphdr *th; 820 struct socket **sop; 821 struct mbuf *m; 822 { 823 struct tcpcb *tp; 824 struct socket *so; 825 struct syncache *sc = NULL; 826 struct syncache_head *sch; 827 struct mbuf *ipopts = NULL; 828 struct rmxp_tao *taop; 829 u_int32_t flowtmp; 830 int i, s, win; 831 832 so = *sop; 833 tp = sototcpcb(so); 834 835 /* 836 * Remember the IP options, if any. 837 */ 838 #ifdef INET6 839 if (!inc->inc_isipv6) 840 #endif 841 ipopts = ip_srcroute(); 842 843 /* 844 * See if we already have an entry for this connection. 845 * If we do, resend the SYN,ACK, and reset the retransmit timer. 846 * 847 * XXX 848 * should the syncache be re-initialized with the contents 849 * of the new SYN here (which may have different options?) 850 */ 851 sc = syncache_lookup(inc, &sch); 852 if (sc != NULL) { 853 tcpstat.tcps_sc_dupsyn++; 854 if (ipopts) { 855 /* 856 * If we were remembering a previous source route, 857 * forget it and use the new one we've been given. 858 */ 859 if (sc->sc_ipopts) 860 (void) m_free(sc->sc_ipopts); 861 sc->sc_ipopts = ipopts; 862 } 863 /* 864 * Update timestamp if present. 865 */ 866 if (sc->sc_flags & SCF_TIMESTAMP) 867 sc->sc_tsrecent = to->to_tsval; 868 /* 869 * PCB may have changed, pick up new values. 870 */ 871 sc->sc_tp = tp; 872 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 873 if (syncache_respond(sc, m) == 0) { 874 s = splnet(); 875 TAILQ_REMOVE(&tcp_syncache.timerq[sc->sc_rxtslot], 876 sc, sc_timerq); 877 SYNCACHE_TIMEOUT(sc, sc->sc_rxtslot); 878 splx(s); 879 tcpstat.tcps_sndacks++; 880 tcpstat.tcps_sndtotal++; 881 } 882 *sop = NULL; 883 return (1); 884 } 885 886 sc = zalloc(tcp_syncache.zone); 887 if (sc == NULL) { 888 /* 889 * The zone allocator couldn't provide more entries. 890 * Treat this as if the cache was full; drop the oldest 891 * entry and insert the new one. 892 */ 893 s = splnet(); 894 for (i = SYNCACHE_MAXREXMTS; i >= 0; i--) { 895 sc = TAILQ_FIRST(&tcp_syncache.timerq[i]); 896 if (sc != NULL) 897 break; 898 } 899 sc->sc_tp->ts_recent = ticks; 900 syncache_drop(sc, NULL); 901 splx(s); 902 tcpstat.tcps_sc_zonefail++; 903 sc = zalloc(tcp_syncache.zone); 904 if (sc == NULL) { 905 if (ipopts) 906 (void) m_free(ipopts); 907 return (0); 908 } 909 } 910 911 /* 912 * Fill in the syncache values. 913 */ 914 bzero(sc, sizeof(*sc)); 915 sc->sc_tp = tp; 916 sc->sc_inp_gencnt = tp->t_inpcb->inp_gencnt; 917 sc->sc_ipopts = ipopts; 918 sc->sc_inc.inc_fport = inc->inc_fport; 919 sc->sc_inc.inc_lport = inc->inc_lport; 920 #ifdef INET6 921 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 922 if (inc->inc_isipv6) { 923 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 924 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 925 sc->sc_route6.ro_rt = NULL; 926 } else 927 #endif 928 { 929 sc->sc_inc.inc_faddr = inc->inc_faddr; 930 sc->sc_inc.inc_laddr = inc->inc_laddr; 931 sc->sc_route.ro_rt = NULL; 932 } 933 sc->sc_irs = th->th_seq; 934 sc->sc_flags = 0; 935 sc->sc_peer_mss = to->to_flags & TOF_MSS ? to->to_mss : 0; 936 sc->sc_flowlabel = 0; 937 if (tcp_syncookies) { 938 sc->sc_iss = syncookie_generate(sc, &flowtmp); 939 #ifdef INET6 940 if (inc->inc_isipv6 && 941 (sc->sc_tp->t_inpcb->in6p_flags & IN6P_AUTOFLOWLABEL)) { 942 sc->sc_flowlabel = flowtmp & IPV6_FLOWLABEL_MASK; 943 } 944 #endif 945 } else { 946 sc->sc_iss = arc4random(); 947 #ifdef INET6 948 if (inc->inc_isipv6 && 949 (sc->sc_tp->t_inpcb->in6p_flags & IN6P_AUTOFLOWLABEL)) { 950 sc->sc_flowlabel = 951 (htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK); 952 } 953 #endif 954 } 955 956 /* Initial receive window: clip sbspace to [0 .. TCP_MAXWIN] */ 957 win = sbspace(&so->so_rcv); 958 win = imax(win, 0); 959 win = imin(win, TCP_MAXWIN); 960 sc->sc_wnd = win; 961 962 if (tcp_do_rfc1323) { 963 /* 964 * A timestamp received in a SYN makes 965 * it ok to send timestamp requests and replies. 966 */ 967 if (to->to_flags & TOF_TS) { 968 sc->sc_tsrecent = to->to_tsval; 969 sc->sc_flags |= SCF_TIMESTAMP; 970 } 971 if (to->to_flags & TOF_SCALE) { 972 int wscale = 0; 973 974 /* Compute proper scaling value from buffer space */ 975 while (wscale < TCP_MAX_WINSHIFT && 976 (TCP_MAXWIN << wscale) < so->so_rcv.sb_hiwat) 977 wscale++; 978 sc->sc_request_r_scale = wscale; 979 sc->sc_requested_s_scale = to->to_requested_s_scale; 980 sc->sc_flags |= SCF_WINSCALE; 981 } 982 } 983 if (tcp_do_rfc1644) { 984 /* 985 * A CC or CC.new option received in a SYN makes 986 * it ok to send CC in subsequent segments. 987 */ 988 if (to->to_flags & (TOF_CC|TOF_CCNEW)) { 989 sc->sc_cc_recv = to->to_cc; 990 sc->sc_cc_send = CC_INC(tcp_ccgen); 991 sc->sc_flags |= SCF_CC; 992 } 993 } 994 if (tp->t_flags & TF_NOOPT) 995 sc->sc_flags = SCF_NOOPT; 996 #ifdef TCP_SIGNATURE 997 /* 998 * If listening socket requested TCP digests, and received SYN 999 * contains the option, flag this in the syncache so that 1000 * syncache_respond() will do the right thing with the SYN+ACK. 1001 * XXX Currently we always record the option by default and will 1002 * attempt to use it in syncache_respond(). 1003 */ 1004 if (to->to_flags & TOF_SIGNATURE) 1005 sc->sc_flags = SCF_SIGNATURE; 1006 #endif 1007 1008 /* 1009 * XXX 1010 * We have the option here of not doing TAO (even if the segment 1011 * qualifies) and instead fall back to a normal 3WHS via the syncache. 1012 * This allows us to apply synflood protection to TAO-qualifying SYNs 1013 * also. However, there should be a hueristic to determine when to 1014 * do this, and is not present at the moment. 1015 */ 1016 1017 /* 1018 * Perform TAO test on incoming CC (SEG.CC) option, if any. 1019 * - compare SEG.CC against cached CC from the same host, if any. 1020 * - if SEG.CC > chached value, SYN must be new and is accepted 1021 * immediately: save new CC in the cache, mark the socket 1022 * connected, enter ESTABLISHED state, turn on flag to 1023 * send a SYN in the next segment. 1024 * A virtual advertised window is set in rcv_adv to 1025 * initialize SWS prevention. Then enter normal segment 1026 * processing: drop SYN, process data and FIN. 1027 * - otherwise do a normal 3-way handshake. 1028 */ 1029 taop = tcp_gettaocache(&sc->sc_inc); 1030 if ((to->to_flags & TOF_CC) != 0) { 1031 if (((tp->t_flags & TF_NOPUSH) != 0) && 1032 sc->sc_flags & SCF_CC && 1033 taop != NULL && taop->tao_cc != 0 && 1034 CC_GT(to->to_cc, taop->tao_cc)) { 1035 sc->sc_rxtslot = 0; 1036 so = syncache_socket(sc, *sop); 1037 if (so != NULL) { 1038 sc->sc_flags |= SCF_KEEPROUTE; 1039 taop->tao_cc = to->to_cc; 1040 *sop = so; 1041 } 1042 syncache_free(sc); 1043 return (so != NULL); 1044 } 1045 } else { 1046 /* 1047 * No CC option, but maybe CC.NEW: invalidate cached value. 1048 */ 1049 if (taop != NULL) 1050 taop->tao_cc = 0; 1051 } 1052 /* 1053 * TAO test failed or there was no CC option, 1054 * do a standard 3-way handshake. 1055 */ 1056 if (syncache_respond(sc, m) == 0) { 1057 syncache_insert(sc, sch); 1058 tcpstat.tcps_sndacks++; 1059 tcpstat.tcps_sndtotal++; 1060 } else { 1061 syncache_free(sc); 1062 tcpstat.tcps_sc_dropped++; 1063 } 1064 *sop = NULL; 1065 return (1); 1066 } 1067 1068 static int 1069 syncache_respond(sc, m) 1070 struct syncache *sc; 1071 struct mbuf *m; 1072 { 1073 u_int8_t *optp; 1074 int optlen, error; 1075 u_int16_t tlen, hlen, mssopt; 1076 struct ip *ip = NULL; 1077 struct rtentry *rt; 1078 struct tcphdr *th; 1079 #ifdef INET6 1080 struct ip6_hdr *ip6 = NULL; 1081 #endif 1082 1083 #ifdef INET6 1084 if (sc->sc_inc.inc_isipv6) { 1085 rt = tcp_rtlookup6(&sc->sc_inc); 1086 if (rt != NULL) 1087 mssopt = rt->rt_ifp->if_mtu - 1088 (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)); 1089 else 1090 mssopt = tcp_v6mssdflt; 1091 hlen = sizeof(struct ip6_hdr); 1092 } else 1093 #endif 1094 { 1095 rt = tcp_rtlookup(&sc->sc_inc); 1096 if (rt != NULL) 1097 mssopt = rt->rt_ifp->if_mtu - 1098 (sizeof(struct ip) + sizeof(struct tcphdr)); 1099 else 1100 mssopt = tcp_mssdflt; 1101 hlen = sizeof(struct ip); 1102 } 1103 1104 /* Compute the size of the TCP options. */ 1105 if (sc->sc_flags & SCF_NOOPT) { 1106 optlen = 0; 1107 } else { 1108 optlen = TCPOLEN_MAXSEG + 1109 ((sc->sc_flags & SCF_WINSCALE) ? 4 : 0) + 1110 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0) + 1111 ((sc->sc_flags & SCF_CC) ? TCPOLEN_CC_APPA * 2 : 0); 1112 #ifdef TCP_SIGNATURE 1113 optlen += ((sc->sc_flags & SCF_SIGNATURE) ? 1114 (TCPOLEN_SIGNATURE + 2) : 0); 1115 #endif 1116 } 1117 tlen = hlen + sizeof(struct tcphdr) + optlen; 1118 1119 /* 1120 * XXX 1121 * assume that the entire packet will fit in a header mbuf 1122 */ 1123 KASSERT(max_linkhdr + tlen <= MHLEN, ("syncache: mbuf too small")); 1124 1125 /* 1126 * XXX shouldn't this reuse the mbuf if possible ? 1127 * Create the IP+TCP header from scratch. 1128 */ 1129 if (m) 1130 m_freem(m); 1131 1132 m = m_gethdr(M_DONTWAIT, MT_HEADER); 1133 if (m == NULL) 1134 return (ENOBUFS); 1135 m->m_data += max_linkhdr; 1136 m->m_len = tlen; 1137 m->m_pkthdr.len = tlen; 1138 m->m_pkthdr.rcvif = NULL; 1139 1140 #ifdef INET6 1141 if (sc->sc_inc.inc_isipv6) { 1142 ip6 = mtod(m, struct ip6_hdr *); 1143 ip6->ip6_vfc = IPV6_VERSION; 1144 ip6->ip6_nxt = IPPROTO_TCP; 1145 ip6->ip6_src = sc->sc_inc.inc6_laddr; 1146 ip6->ip6_dst = sc->sc_inc.inc6_faddr; 1147 ip6->ip6_plen = htons(tlen - hlen); 1148 /* ip6_hlim is set after checksum */ 1149 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; 1150 ip6->ip6_flow |= sc->sc_flowlabel; 1151 1152 th = (struct tcphdr *)(ip6 + 1); 1153 } else 1154 #endif 1155 { 1156 ip = mtod(m, struct ip *); 1157 ip->ip_v = IPVERSION; 1158 ip->ip_hl = sizeof(struct ip) >> 2; 1159 ip->ip_len = tlen; 1160 ip->ip_id = 0; 1161 ip->ip_off = 0; 1162 ip->ip_sum = 0; 1163 ip->ip_p = IPPROTO_TCP; 1164 ip->ip_src = sc->sc_inc.inc_laddr; 1165 ip->ip_dst = sc->sc_inc.inc_faddr; 1166 ip->ip_ttl = sc->sc_tp->t_inpcb->inp_ip_ttl; /* XXX */ 1167 ip->ip_tos = sc->sc_tp->t_inpcb->inp_ip_tos; /* XXX */ 1168 1169 /* 1170 * See if we should do MTU discovery. Route lookups are expensive, 1171 * so we will only unset the DF bit if: 1172 * 1173 * 1) path_mtu_discovery is disabled 1174 * 2) the SCF_UNREACH flag has been set 1175 */ 1176 if (path_mtu_discovery 1177 && ((sc->sc_flags & SCF_UNREACH) == 0)) { 1178 ip->ip_off |= IP_DF; 1179 } 1180 1181 th = (struct tcphdr *)(ip + 1); 1182 } 1183 th->th_sport = sc->sc_inc.inc_lport; 1184 th->th_dport = sc->sc_inc.inc_fport; 1185 1186 th->th_seq = htonl(sc->sc_iss); 1187 th->th_ack = htonl(sc->sc_irs + 1); 1188 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 1189 th->th_x2 = 0; 1190 th->th_flags = TH_SYN|TH_ACK; 1191 th->th_win = htons(sc->sc_wnd); 1192 th->th_urp = 0; 1193 1194 /* Tack on the TCP options. */ 1195 if (optlen == 0) 1196 goto no_options; 1197 optp = (u_int8_t *)(th + 1); 1198 *optp++ = TCPOPT_MAXSEG; 1199 *optp++ = TCPOLEN_MAXSEG; 1200 *optp++ = (mssopt >> 8) & 0xff; 1201 *optp++ = mssopt & 0xff; 1202 1203 if (sc->sc_flags & SCF_WINSCALE) { 1204 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 1205 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 1206 sc->sc_request_r_scale); 1207 optp += 4; 1208 } 1209 1210 if (sc->sc_flags & SCF_TIMESTAMP) { 1211 u_int32_t *lp = (u_int32_t *)(optp); 1212 1213 /* Form timestamp option as shown in appendix A of RFC 1323. */ 1214 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 1215 *lp++ = htonl(ticks); 1216 *lp = htonl(sc->sc_tsrecent); 1217 optp += TCPOLEN_TSTAMP_APPA; 1218 } 1219 1220 /* 1221 * Send CC and CC.echo if we received CC from our peer. 1222 */ 1223 if (sc->sc_flags & SCF_CC) { 1224 u_int32_t *lp = (u_int32_t *)(optp); 1225 1226 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CC)); 1227 *lp++ = htonl(sc->sc_cc_send); 1228 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CCECHO)); 1229 *lp = htonl(sc->sc_cc_recv); 1230 optp += TCPOLEN_CC_APPA * 2; 1231 } 1232 #ifdef TCP_SIGNATURE 1233 /* 1234 * Handle TCP-MD5 passive opener response. 1235 */ 1236 if (sc->sc_flags & SCF_SIGNATURE) { 1237 u_int8_t *bp = optp; 1238 int i; 1239 1240 *bp++ = TCPOPT_SIGNATURE; 1241 *bp++ = TCPOLEN_SIGNATURE; 1242 for (i = 0; i < TCP_SIGLEN; i++) 1243 *bp++ = 0; 1244 tcp_signature_compute(m, sizeof(struct ip), 0, optlen, 1245 optp + 2, IPSEC_DIR_OUTBOUND); 1246 *bp++ = TCPOPT_NOP; 1247 *bp++ = TCPOPT_EOL; 1248 optp += TCPOLEN_SIGNATURE + 2; 1249 } 1250 #endif /* TCP_SIGNATURE */ 1251 no_options: 1252 1253 #ifdef INET6 1254 if (sc->sc_inc.inc_isipv6) { 1255 struct route_in6 *ro6 = &sc->sc_route6; 1256 1257 th->th_sum = 0; 1258 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 1259 ip6->ip6_hlim = in6_selecthlim(NULL, 1260 ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL); 1261 error = ip6_output(m, NULL, ro6, 0, NULL, NULL, 1262 sc->sc_tp->t_inpcb); 1263 } else 1264 #endif 1265 { 1266 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1267 htons(tlen - hlen + IPPROTO_TCP)); 1268 m->m_pkthdr.csum_flags = CSUM_TCP; 1269 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1270 error = ip_output(m, sc->sc_ipopts, &sc->sc_route, 0, NULL, 1271 sc->sc_tp->t_inpcb); 1272 } 1273 return (error); 1274 } 1275 1276 /* 1277 * cookie layers: 1278 * 1279 * |. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .| 1280 * | peer iss | 1281 * | MD5(laddr,faddr,secret,lport,fport) |. . . . . . .| 1282 * | 0 |(A)| | 1283 * (A): peer mss index 1284 */ 1285 1286 /* 1287 * The values below are chosen to minimize the size of the tcp_secret 1288 * table, as well as providing roughly a 16 second lifetime for the cookie. 1289 */ 1290 1291 #define SYNCOOKIE_WNDBITS 5 /* exposed bits for window indexing */ 1292 #define SYNCOOKIE_TIMESHIFT 1 /* scale ticks to window time units */ 1293 1294 #define SYNCOOKIE_WNDMASK ((1 << SYNCOOKIE_WNDBITS) - 1) 1295 #define SYNCOOKIE_NSECRETS (1 << SYNCOOKIE_WNDBITS) 1296 #define SYNCOOKIE_TIMEOUT \ 1297 (hz * (1 << SYNCOOKIE_WNDBITS) / (1 << SYNCOOKIE_TIMESHIFT)) 1298 #define SYNCOOKIE_DATAMASK ((3 << SYNCOOKIE_WNDBITS) | SYNCOOKIE_WNDMASK) 1299 1300 static struct { 1301 u_int32_t ts_secbits[4]; 1302 u_int ts_expire; 1303 } tcp_secret[SYNCOOKIE_NSECRETS]; 1304 1305 static int tcp_msstab[] = { 0, 536, 1460, 8960 }; 1306 1307 static MD5_CTX syn_ctx; 1308 1309 #define MD5Add(v) MD5Update(&syn_ctx, (u_char *)&v, sizeof(v)) 1310 1311 struct md5_add { 1312 u_int32_t laddr, faddr; 1313 u_int32_t secbits[4]; 1314 u_int16_t lport, fport; 1315 }; 1316 1317 #ifdef CTASSERT 1318 CTASSERT(sizeof(struct md5_add) == 28); 1319 #endif 1320 1321 /* 1322 * Consider the problem of a recreated (and retransmitted) cookie. If the 1323 * original SYN was accepted, the connection is established. The second 1324 * SYN is inflight, and if it arrives with an ISN that falls within the 1325 * receive window, the connection is killed. 1326 * 1327 * However, since cookies have other problems, this may not be worth 1328 * worrying about. 1329 */ 1330 1331 static u_int32_t 1332 syncookie_generate(struct syncache *sc, u_int32_t *flowid) 1333 { 1334 u_int32_t md5_buffer[4]; 1335 u_int32_t data; 1336 int idx, i; 1337 struct md5_add add; 1338 1339 idx = ((ticks << SYNCOOKIE_TIMESHIFT) / hz) & SYNCOOKIE_WNDMASK; 1340 if (tcp_secret[idx].ts_expire < ticks) { 1341 for (i = 0; i < 4; i++) 1342 tcp_secret[idx].ts_secbits[i] = arc4random(); 1343 tcp_secret[idx].ts_expire = ticks + SYNCOOKIE_TIMEOUT; 1344 } 1345 for (data = sizeof(tcp_msstab) / sizeof(int) - 1; data > 0; data--) 1346 if (tcp_msstab[data] <= sc->sc_peer_mss) 1347 break; 1348 data = (data << SYNCOOKIE_WNDBITS) | idx; 1349 data ^= sc->sc_irs; /* peer's iss */ 1350 MD5Init(&syn_ctx); 1351 #ifdef INET6 1352 if (sc->sc_inc.inc_isipv6) { 1353 MD5Add(sc->sc_inc.inc6_laddr); 1354 MD5Add(sc->sc_inc.inc6_faddr); 1355 add.laddr = 0; 1356 add.faddr = 0; 1357 } else 1358 #endif 1359 { 1360 add.laddr = sc->sc_inc.inc_laddr.s_addr; 1361 add.faddr = sc->sc_inc.inc_faddr.s_addr; 1362 } 1363 add.lport = sc->sc_inc.inc_lport; 1364 add.fport = sc->sc_inc.inc_fport; 1365 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1366 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1367 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1368 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1369 MD5Add(add); 1370 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1371 data ^= (md5_buffer[0] & ~SYNCOOKIE_WNDMASK); 1372 *flowid = md5_buffer[1]; 1373 return (data); 1374 } 1375 1376 static struct syncache * 1377 syncookie_lookup(inc, th, so) 1378 struct in_conninfo *inc; 1379 struct tcphdr *th; 1380 struct socket *so; 1381 { 1382 u_int32_t md5_buffer[4]; 1383 struct syncache *sc; 1384 u_int32_t data; 1385 int wnd, idx; 1386 struct md5_add add; 1387 1388 data = (th->th_ack - 1) ^ (th->th_seq - 1); /* remove ISS */ 1389 idx = data & SYNCOOKIE_WNDMASK; 1390 if (tcp_secret[idx].ts_expire < ticks || 1391 sototcpcb(so)->ts_recent + SYNCOOKIE_TIMEOUT < ticks) 1392 return (NULL); 1393 MD5Init(&syn_ctx); 1394 #ifdef INET6 1395 if (inc->inc_isipv6) { 1396 MD5Add(inc->inc6_laddr); 1397 MD5Add(inc->inc6_faddr); 1398 add.laddr = 0; 1399 add.faddr = 0; 1400 } else 1401 #endif 1402 { 1403 add.laddr = inc->inc_laddr.s_addr; 1404 add.faddr = inc->inc_faddr.s_addr; 1405 } 1406 add.lport = inc->inc_lport; 1407 add.fport = inc->inc_fport; 1408 add.secbits[0] = tcp_secret[idx].ts_secbits[0]; 1409 add.secbits[1] = tcp_secret[idx].ts_secbits[1]; 1410 add.secbits[2] = tcp_secret[idx].ts_secbits[2]; 1411 add.secbits[3] = tcp_secret[idx].ts_secbits[3]; 1412 MD5Add(add); 1413 MD5Final((u_char *)&md5_buffer, &syn_ctx); 1414 data ^= md5_buffer[0]; 1415 if ((data & ~SYNCOOKIE_DATAMASK) != 0) 1416 return (NULL); 1417 data = data >> SYNCOOKIE_WNDBITS; 1418 1419 sc = zalloc(tcp_syncache.zone); 1420 if (sc == NULL) 1421 return (NULL); 1422 /* 1423 * Fill in the syncache values. 1424 * XXX duplicate code from syncache_add 1425 */ 1426 sc->sc_ipopts = NULL; 1427 sc->sc_inc.inc_fport = inc->inc_fport; 1428 sc->sc_inc.inc_lport = inc->inc_lport; 1429 sc->sc_tp = sototcpcb(so); 1430 #ifdef INET6 1431 sc->sc_inc.inc_isipv6 = inc->inc_isipv6; 1432 if (inc->inc_isipv6) { 1433 sc->sc_inc.inc6_faddr = inc->inc6_faddr; 1434 sc->sc_inc.inc6_laddr = inc->inc6_laddr; 1435 sc->sc_route6.ro_rt = NULL; 1436 if (sc->sc_tp->t_inpcb->in6p_flags & IN6P_AUTOFLOWLABEL) 1437 sc->sc_flowlabel = md5_buffer[1] & IPV6_FLOWLABEL_MASK; 1438 } else 1439 #endif 1440 { 1441 sc->sc_inc.inc_faddr = inc->inc_faddr; 1442 sc->sc_inc.inc_laddr = inc->inc_laddr; 1443 sc->sc_route.ro_rt = NULL; 1444 } 1445 sc->sc_irs = th->th_seq - 1; 1446 sc->sc_iss = th->th_ack - 1; 1447 wnd = sbspace(&so->so_rcv); 1448 wnd = imax(wnd, 0); 1449 wnd = imin(wnd, TCP_MAXWIN); 1450 sc->sc_wnd = wnd; 1451 sc->sc_flags = 0; 1452 sc->sc_rxtslot = 0; 1453 sc->sc_peer_mss = tcp_msstab[data]; 1454 return (sc); 1455 }
Cache object: c39e74f232c96d22efaead190208f304
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]